1 // SPDX-License-Identifier: GPL-2.0-only 2 /* 3 * kernel/sched/debug.c 4 * 5 * Print the CFS rbtree and other debugging details 6 * 7 * Copyright(C) 2007, Red Hat, Inc., Ingo Molnar 8 */ 9 #include <linux/debugfs.h> 10 #include <linux/nmi.h> 11 #include <linux/log2.h> 12 #include "sched.h" 13 14 /* 15 * This allows printing both to /sys/kernel/debug/sched/debug and 16 * to the console 17 */ 18 #define SEQ_printf(m, x...) \ 19 do { \ 20 if (m) \ 21 seq_printf(m, x); \ 22 else \ 23 pr_cont(x); \ 24 } while (0) 25 26 /* 27 * Ease the printing of nsec fields: 28 */ 29 static long long nsec_high(unsigned long long nsec) 30 { 31 if ((long long)nsec < 0) { 32 nsec = -nsec; 33 do_div(nsec, 1000000); 34 return -nsec; 35 } 36 do_div(nsec, 1000000); 37 38 return nsec; 39 } 40 41 static unsigned long nsec_low(unsigned long long nsec) 42 { 43 if ((long long)nsec < 0) 44 nsec = -nsec; 45 46 return do_div(nsec, 1000000); 47 } 48 49 #define SPLIT_NS(x) nsec_high(x), nsec_low(x) 50 51 #define SCHED_FEAT(name, enabled) \ 52 #name , 53 54 static const char * const sched_feat_names[] = { 55 #include "features.h" 56 }; 57 58 #undef SCHED_FEAT 59 60 static int sched_feat_show(struct seq_file *m, void *v) 61 { 62 int i; 63 64 for (i = 0; i < __SCHED_FEAT_NR; i++) { 65 if (!(sysctl_sched_features & (1UL << i))) 66 seq_puts(m, "NO_"); 67 seq_printf(m, "%s ", sched_feat_names[i]); 68 } 69 seq_puts(m, "\n"); 70 71 return 0; 72 } 73 74 #ifdef CONFIG_JUMP_LABEL 75 76 #define jump_label_key__true STATIC_KEY_INIT_TRUE 77 #define jump_label_key__false STATIC_KEY_INIT_FALSE 78 79 #define SCHED_FEAT(name, enabled) \ 80 jump_label_key__##enabled , 81 82 struct static_key sched_feat_keys[__SCHED_FEAT_NR] = { 83 #include "features.h" 84 }; 85 86 #undef SCHED_FEAT 87 88 static void sched_feat_disable(int i) 89 { 90 static_key_disable_cpuslocked(&sched_feat_keys[i]); 91 } 92 93 static void sched_feat_enable(int i) 94 { 95 static_key_enable_cpuslocked(&sched_feat_keys[i]); 96 } 97 #else /* !CONFIG_JUMP_LABEL: */ 98 static void sched_feat_disable(int i) { }; 99 static void sched_feat_enable(int i) { }; 100 #endif /* !CONFIG_JUMP_LABEL */ 101 102 static int sched_feat_set(char *cmp) 103 { 104 int i; 105 int neg = 0; 106 107 if (strncmp(cmp, "NO_", 3) == 0) { 108 neg = 1; 109 cmp += 3; 110 } 111 112 i = match_string(sched_feat_names, __SCHED_FEAT_NR, cmp); 113 if (i < 0) 114 return i; 115 116 if (neg) { 117 sysctl_sched_features &= ~(1UL << i); 118 sched_feat_disable(i); 119 } else { 120 sysctl_sched_features |= (1UL << i); 121 sched_feat_enable(i); 122 } 123 124 return 0; 125 } 126 127 static ssize_t 128 sched_feat_write(struct file *filp, const char __user *ubuf, 129 size_t cnt, loff_t *ppos) 130 { 131 char buf[64]; 132 char *cmp; 133 int ret; 134 struct inode *inode; 135 136 if (cnt > 63) 137 cnt = 63; 138 139 if (copy_from_user(buf, ubuf, cnt)) 140 return -EFAULT; 141 142 buf[cnt] = 0; 143 cmp = strstrip(buf); 144 145 /* Ensure the static_key remains in a consistent state */ 146 inode = file_inode(filp); 147 cpus_read_lock(); 148 inode_lock(inode); 149 ret = sched_feat_set(cmp); 150 inode_unlock(inode); 151 cpus_read_unlock(); 152 if (ret < 0) 153 return ret; 154 155 *ppos += cnt; 156 157 return cnt; 158 } 159 160 static int sched_feat_open(struct inode *inode, struct file *filp) 161 { 162 return single_open(filp, sched_feat_show, NULL); 163 } 164 165 static const struct file_operations sched_feat_fops = { 166 .open = sched_feat_open, 167 .write = sched_feat_write, 168 .read = seq_read, 169 .llseek = seq_lseek, 170 .release = single_release, 171 }; 172 173 static ssize_t sched_scaling_write(struct file *filp, const char __user *ubuf, 174 size_t cnt, loff_t *ppos) 175 { 176 unsigned int scaling; 177 int ret; 178 179 ret = kstrtouint_from_user(ubuf, cnt, 10, &scaling); 180 if (ret) 181 return ret; 182 183 if (scaling >= SCHED_TUNABLESCALING_END) 184 return -EINVAL; 185 186 sysctl_sched_tunable_scaling = scaling; 187 if (sched_update_scaling()) 188 return -EINVAL; 189 190 *ppos += cnt; 191 return cnt; 192 } 193 194 static int sched_scaling_show(struct seq_file *m, void *v) 195 { 196 seq_printf(m, "%d\n", sysctl_sched_tunable_scaling); 197 return 0; 198 } 199 200 static int sched_scaling_open(struct inode *inode, struct file *filp) 201 { 202 return single_open(filp, sched_scaling_show, NULL); 203 } 204 205 static const struct file_operations sched_scaling_fops = { 206 .open = sched_scaling_open, 207 .write = sched_scaling_write, 208 .read = seq_read, 209 .llseek = seq_lseek, 210 .release = single_release, 211 }; 212 213 #ifdef CONFIG_SCHED_CACHE 214 static ssize_t 215 sched_cache_enable_write(struct file *filp, const char __user *ubuf, 216 size_t cnt, loff_t *ppos) 217 { 218 bool val; 219 int ret; 220 221 ret = kstrtobool_from_user(ubuf, cnt, &val); 222 if (ret) 223 return ret; 224 225 sysctl_sched_cache_user = val; 226 227 sched_cache_active_set(); 228 229 *ppos += cnt; 230 231 return cnt; 232 } 233 234 static int sched_cache_enable_show(struct seq_file *m, void *v) 235 { 236 seq_printf(m, "%d\n", sysctl_sched_cache_user); 237 return 0; 238 } 239 240 static int sched_cache_enable_open(struct inode *inode, 241 struct file *filp) 242 { 243 return single_open(filp, sched_cache_enable_show, NULL); 244 } 245 246 static const struct file_operations sched_cache_enable_fops = { 247 .open = sched_cache_enable_open, 248 .write = sched_cache_enable_write, 249 .read = seq_read, 250 .llseek = seq_lseek, 251 .release = single_release, 252 }; 253 #endif 254 255 #ifdef CONFIG_PREEMPT_DYNAMIC 256 257 static ssize_t sched_dynamic_write(struct file *filp, const char __user *ubuf, 258 size_t cnt, loff_t *ppos) 259 { 260 char buf[16]; 261 int mode; 262 263 if (cnt > 15) 264 cnt = 15; 265 266 if (copy_from_user(buf, ubuf, cnt)) 267 return -EFAULT; 268 269 buf[cnt] = 0; 270 mode = sched_dynamic_mode(strstrip(buf)); 271 if (mode < 0) 272 return mode; 273 274 sched_dynamic_update(mode); 275 276 *ppos += cnt; 277 278 return cnt; 279 } 280 281 static int sched_dynamic_show(struct seq_file *m, void *v) 282 { 283 int i = (IS_ENABLED(CONFIG_PREEMPT_RT) || IS_ENABLED(CONFIG_ARCH_HAS_PREEMPT_LAZY)) * 2; 284 int mode = READ_ONCE(preempt_dynamic_mode); 285 int j; 286 287 /* Count entries in NULL terminated preempt_modes */ 288 for (j = 0; preempt_modes[j]; j++) 289 ; 290 j -= !IS_ENABLED(CONFIG_ARCH_HAS_PREEMPT_LAZY); 291 292 for (; i < j; i++) { 293 if (mode == i) 294 seq_puts(m, "("); 295 seq_puts(m, preempt_modes[i]); 296 if (mode == i) 297 seq_puts(m, ")"); 298 299 seq_puts(m, " "); 300 } 301 302 seq_puts(m, "\n"); 303 return 0; 304 } 305 306 static int sched_dynamic_open(struct inode *inode, struct file *filp) 307 { 308 return single_open(filp, sched_dynamic_show, NULL); 309 } 310 311 static const struct file_operations sched_dynamic_fops = { 312 .open = sched_dynamic_open, 313 .write = sched_dynamic_write, 314 .read = seq_read, 315 .llseek = seq_lseek, 316 .release = single_release, 317 }; 318 319 #endif /* CONFIG_PREEMPT_DYNAMIC */ 320 321 __read_mostly bool sched_debug_verbose; 322 323 static struct dentry *sd_dentry; 324 325 326 static ssize_t sched_verbose_write(struct file *filp, const char __user *ubuf, 327 size_t cnt, loff_t *ppos) 328 { 329 ssize_t result; 330 bool orig; 331 332 cpus_read_lock(); 333 sched_domains_mutex_lock(); 334 335 orig = sched_debug_verbose; 336 result = debugfs_write_file_bool(filp, ubuf, cnt, ppos); 337 338 if (sched_debug_verbose && !orig) 339 update_sched_domain_debugfs(); 340 else if (!sched_debug_verbose && orig) { 341 debugfs_remove(sd_dentry); 342 sd_dentry = NULL; 343 } 344 345 sched_domains_mutex_unlock(); 346 cpus_read_unlock(); 347 348 return result; 349 } 350 351 static const struct file_operations sched_verbose_fops = { 352 .read = debugfs_read_file_bool, 353 .write = sched_verbose_write, 354 .open = simple_open, 355 .llseek = default_llseek, 356 }; 357 358 static const struct seq_operations sched_debug_sops; 359 360 static int sched_debug_open(struct inode *inode, struct file *filp) 361 { 362 return seq_open(filp, &sched_debug_sops); 363 } 364 365 static const struct file_operations sched_debug_fops = { 366 .open = sched_debug_open, 367 .read = seq_read, 368 .llseek = seq_lseek, 369 .release = seq_release, 370 }; 371 372 enum dl_param { 373 DL_RUNTIME = 0, 374 DL_PERIOD, 375 }; 376 377 static unsigned long dl_server_period_max = (1UL << 22) * NSEC_PER_USEC; /* ~4 seconds */ 378 static unsigned long dl_server_period_min = (100) * NSEC_PER_USEC; /* 100 us */ 379 380 static ssize_t sched_server_write_common(struct file *filp, const char __user *ubuf, 381 size_t cnt, loff_t *ppos, enum dl_param param, 382 void *server) 383 { 384 long cpu = (long) ((struct seq_file *) filp->private_data)->private; 385 struct sched_dl_entity *dl_se = (struct sched_dl_entity *)server; 386 u64 old_runtime, runtime, period; 387 struct rq *rq = cpu_rq(cpu); 388 int retval = 0; 389 size_t err; 390 u64 value; 391 392 err = kstrtoull_from_user(ubuf, cnt, 10, &value); 393 if (err) 394 return err; 395 396 scoped_guard (rq_lock_irqsave, rq) { 397 old_runtime = runtime = dl_se->dl_runtime; 398 period = dl_se->dl_period; 399 400 switch (param) { 401 case DL_RUNTIME: 402 if (runtime == value) 403 break; 404 runtime = value; 405 break; 406 case DL_PERIOD: 407 if (value == period) 408 break; 409 period = value; 410 break; 411 } 412 413 if (runtime > period || 414 period > dl_server_period_max || 415 period < dl_server_period_min) { 416 return -EINVAL; 417 } 418 419 if (!cpu_online(cpu_of(rq))) 420 return -EBUSY; 421 422 update_rq_clock(rq); 423 dl_server_stop(dl_se); 424 retval = dl_server_apply_params(dl_se, runtime, period, 0); 425 dl_server_start(dl_se); 426 427 if (retval < 0) 428 return retval; 429 } 430 431 if (!!old_runtime ^ !!runtime) { 432 pr_info("%s server %sabled on CPU %d%s.\n", 433 server == &rq->fair_server ? "Fair" : "Ext", 434 runtime ? "en" : "dis", 435 cpu_of(rq), 436 runtime ? "" : ", system may malfunction due to starvation"); 437 } 438 439 *ppos += cnt; 440 return cnt; 441 } 442 443 static size_t sched_server_show_common(struct seq_file *m, void *v, enum dl_param param, 444 void *server) 445 { 446 struct sched_dl_entity *dl_se = (struct sched_dl_entity *)server; 447 u64 value; 448 449 switch (param) { 450 case DL_RUNTIME: 451 value = dl_se->dl_runtime; 452 break; 453 case DL_PERIOD: 454 value = dl_se->dl_period; 455 break; 456 } 457 458 seq_printf(m, "%llu\n", value); 459 return 0; 460 } 461 462 static ssize_t 463 sched_fair_server_runtime_write(struct file *filp, const char __user *ubuf, 464 size_t cnt, loff_t *ppos) 465 { 466 long cpu = (long) ((struct seq_file *) filp->private_data)->private; 467 struct rq *rq = cpu_rq(cpu); 468 469 return sched_server_write_common(filp, ubuf, cnt, ppos, DL_RUNTIME, 470 &rq->fair_server); 471 } 472 473 static int sched_fair_server_runtime_show(struct seq_file *m, void *v) 474 { 475 unsigned long cpu = (unsigned long) m->private; 476 struct rq *rq = cpu_rq(cpu); 477 478 return sched_server_show_common(m, v, DL_RUNTIME, &rq->fair_server); 479 } 480 481 static int sched_fair_server_runtime_open(struct inode *inode, struct file *filp) 482 { 483 return single_open(filp, sched_fair_server_runtime_show, inode->i_private); 484 } 485 486 static const struct file_operations fair_server_runtime_fops = { 487 .open = sched_fair_server_runtime_open, 488 .write = sched_fair_server_runtime_write, 489 .read = seq_read, 490 .llseek = seq_lseek, 491 .release = single_release, 492 }; 493 494 static struct dentry *debugfs_sched; 495 496 #ifdef CONFIG_SCHED_CLASS_EXT 497 static ssize_t 498 sched_ext_server_runtime_write(struct file *filp, const char __user *ubuf, 499 size_t cnt, loff_t *ppos) 500 { 501 long cpu = (long) ((struct seq_file *) filp->private_data)->private; 502 struct rq *rq = cpu_rq(cpu); 503 504 return sched_server_write_common(filp, ubuf, cnt, ppos, DL_RUNTIME, 505 &rq->ext_server); 506 } 507 508 static int sched_ext_server_runtime_show(struct seq_file *m, void *v) 509 { 510 unsigned long cpu = (unsigned long) m->private; 511 struct rq *rq = cpu_rq(cpu); 512 513 return sched_server_show_common(m, v, DL_RUNTIME, &rq->ext_server); 514 } 515 516 static int sched_ext_server_runtime_open(struct inode *inode, struct file *filp) 517 { 518 return single_open(filp, sched_ext_server_runtime_show, inode->i_private); 519 } 520 521 static const struct file_operations ext_server_runtime_fops = { 522 .open = sched_ext_server_runtime_open, 523 .write = sched_ext_server_runtime_write, 524 .read = seq_read, 525 .llseek = seq_lseek, 526 .release = single_release, 527 }; 528 529 static ssize_t 530 sched_ext_server_period_write(struct file *filp, const char __user *ubuf, 531 size_t cnt, loff_t *ppos) 532 { 533 long cpu = (long) ((struct seq_file *) filp->private_data)->private; 534 struct rq *rq = cpu_rq(cpu); 535 536 return sched_server_write_common(filp, ubuf, cnt, ppos, DL_PERIOD, 537 &rq->ext_server); 538 } 539 540 static int sched_ext_server_period_show(struct seq_file *m, void *v) 541 { 542 unsigned long cpu = (unsigned long) m->private; 543 struct rq *rq = cpu_rq(cpu); 544 545 return sched_server_show_common(m, v, DL_PERIOD, &rq->ext_server); 546 } 547 548 static int sched_ext_server_period_open(struct inode *inode, struct file *filp) 549 { 550 return single_open(filp, sched_ext_server_period_show, inode->i_private); 551 } 552 553 static const struct file_operations ext_server_period_fops = { 554 .open = sched_ext_server_period_open, 555 .write = sched_ext_server_period_write, 556 .read = seq_read, 557 .llseek = seq_lseek, 558 .release = single_release, 559 }; 560 561 static void debugfs_ext_server_init(void) 562 { 563 struct dentry *d_ext; 564 unsigned long cpu; 565 566 d_ext = debugfs_create_dir("ext_server", debugfs_sched); 567 if (!d_ext) 568 return; 569 570 for_each_possible_cpu(cpu) { 571 struct dentry *d_cpu; 572 char buf[32]; 573 574 snprintf(buf, sizeof(buf), "cpu%lu", cpu); 575 d_cpu = debugfs_create_dir(buf, d_ext); 576 577 debugfs_create_file("runtime", 0644, d_cpu, (void *) cpu, &ext_server_runtime_fops); 578 debugfs_create_file("period", 0644, d_cpu, (void *) cpu, &ext_server_period_fops); 579 } 580 } 581 #endif /* CONFIG_SCHED_CLASS_EXT */ 582 583 static ssize_t 584 sched_fair_server_period_write(struct file *filp, const char __user *ubuf, 585 size_t cnt, loff_t *ppos) 586 { 587 long cpu = (long) ((struct seq_file *) filp->private_data)->private; 588 struct rq *rq = cpu_rq(cpu); 589 590 return sched_server_write_common(filp, ubuf, cnt, ppos, DL_PERIOD, 591 &rq->fair_server); 592 } 593 594 static int sched_fair_server_period_show(struct seq_file *m, void *v) 595 { 596 unsigned long cpu = (unsigned long) m->private; 597 struct rq *rq = cpu_rq(cpu); 598 599 return sched_server_show_common(m, v, DL_PERIOD, &rq->fair_server); 600 } 601 602 static int sched_fair_server_period_open(struct inode *inode, struct file *filp) 603 { 604 return single_open(filp, sched_fair_server_period_show, inode->i_private); 605 } 606 607 static const struct file_operations fair_server_period_fops = { 608 .open = sched_fair_server_period_open, 609 .write = sched_fair_server_period_write, 610 .read = seq_read, 611 .llseek = seq_lseek, 612 .release = single_release, 613 }; 614 615 static void debugfs_fair_server_init(void) 616 { 617 struct dentry *d_fair; 618 unsigned long cpu; 619 620 d_fair = debugfs_create_dir("fair_server", debugfs_sched); 621 if (!d_fair) 622 return; 623 624 for_each_possible_cpu(cpu) { 625 struct dentry *d_cpu; 626 char buf[32]; 627 628 snprintf(buf, sizeof(buf), "cpu%lu", cpu); 629 d_cpu = debugfs_create_dir(buf, d_fair); 630 631 debugfs_create_file("runtime", 0644, d_cpu, (void *) cpu, &fair_server_runtime_fops); 632 debugfs_create_file("period", 0644, d_cpu, (void *) cpu, &fair_server_period_fops); 633 } 634 } 635 636 static __init int sched_init_debug(void) 637 { 638 struct dentry __maybe_unused *numa, *llc; 639 640 debugfs_sched = debugfs_create_dir("sched", NULL); 641 642 debugfs_create_file("features", 0644, debugfs_sched, NULL, &sched_feat_fops); 643 debugfs_create_file_unsafe("verbose", 0644, debugfs_sched, &sched_debug_verbose, &sched_verbose_fops); 644 #ifdef CONFIG_PREEMPT_DYNAMIC 645 debugfs_create_file("preempt", 0644, debugfs_sched, NULL, &sched_dynamic_fops); 646 #endif 647 648 debugfs_create_u32("base_slice_ns", 0644, debugfs_sched, &sysctl_sched_base_slice); 649 650 debugfs_create_u32("latency_warn_ms", 0644, debugfs_sched, &sysctl_resched_latency_warn_ms); 651 debugfs_create_u32("latency_warn_once", 0644, debugfs_sched, &sysctl_resched_latency_warn_once); 652 653 debugfs_create_file("tunable_scaling", 0644, debugfs_sched, NULL, &sched_scaling_fops); 654 debugfs_create_u32("migration_cost_ns", 0644, debugfs_sched, &sysctl_sched_migration_cost); 655 debugfs_create_u32("nr_migrate", 0644, debugfs_sched, &sysctl_sched_nr_migrate); 656 657 sched_domains_mutex_lock(); 658 update_sched_domain_debugfs(); 659 sched_domains_mutex_unlock(); 660 661 #ifdef CONFIG_NUMA_BALANCING 662 numa = debugfs_create_dir("numa_balancing", debugfs_sched); 663 664 debugfs_create_u32("scan_delay_ms", 0644, numa, &sysctl_numa_balancing_scan_delay); 665 debugfs_create_u32("scan_period_min_ms", 0644, numa, &sysctl_numa_balancing_scan_period_min); 666 debugfs_create_u32("scan_period_max_ms", 0644, numa, &sysctl_numa_balancing_scan_period_max); 667 debugfs_create_u32("scan_size_mb", 0644, numa, &sysctl_numa_balancing_scan_size); 668 debugfs_create_u32("hot_threshold_ms", 0644, numa, &sysctl_numa_balancing_hot_threshold); 669 #endif /* CONFIG_NUMA_BALANCING */ 670 671 #ifdef CONFIG_SCHED_CACHE 672 llc = debugfs_create_dir("llc_balancing", debugfs_sched); 673 debugfs_create_file("enabled", 0644, llc, NULL, 674 &sched_cache_enable_fops); 675 debugfs_create_u32("aggr_tolerance", 0644, llc, 676 &llc_aggr_tolerance); 677 debugfs_create_u32("epoch_period", 0644, llc, 678 &llc_epoch_period); 679 debugfs_create_u32("epoch_affinity_timeout", 0644, llc, 680 &llc_epoch_affinity_timeout); 681 debugfs_create_u32("overaggr_pct", 0644, llc, 682 &llc_overaggr_pct); 683 debugfs_create_u32("imb_pct", 0644, llc, 684 &llc_imb_pct); 685 #endif 686 687 debugfs_create_file("debug", 0444, debugfs_sched, NULL, &sched_debug_fops); 688 689 debugfs_fair_server_init(); 690 #ifdef CONFIG_SCHED_CLASS_EXT 691 debugfs_ext_server_init(); 692 #endif 693 694 return 0; 695 } 696 late_initcall(sched_init_debug); 697 698 static cpumask_var_t sd_sysctl_cpus; 699 700 static int sd_flags_show(struct seq_file *m, void *v) 701 { 702 unsigned long flags = *(unsigned int *)m->private; 703 int idx; 704 705 for_each_set_bit(idx, &flags, __SD_FLAG_CNT) { 706 seq_puts(m, sd_flag_debug[idx].name); 707 seq_puts(m, " "); 708 } 709 seq_puts(m, "\n"); 710 711 return 0; 712 } 713 714 static int sd_flags_open(struct inode *inode, struct file *file) 715 { 716 return single_open(file, sd_flags_show, inode->i_private); 717 } 718 719 static const struct file_operations sd_flags_fops = { 720 .open = sd_flags_open, 721 .read = seq_read, 722 .llseek = seq_lseek, 723 .release = single_release, 724 }; 725 726 static void register_sd(struct sched_domain *sd, struct dentry *parent) 727 { 728 #define SDM(type, mode, member) \ 729 debugfs_create_##type(#member, mode, parent, &sd->member) 730 731 SDM(ulong, 0644, min_interval); 732 SDM(ulong, 0644, max_interval); 733 SDM(u64, 0644, max_newidle_lb_cost); 734 SDM(u32, 0644, busy_factor); 735 SDM(u32, 0644, imbalance_pct); 736 SDM(u32, 0644, cache_nice_tries); 737 SDM(str, 0444, name); 738 739 #undef SDM 740 741 debugfs_create_file("flags", 0444, parent, &sd->flags, &sd_flags_fops); 742 debugfs_create_file("groups_flags", 0444, parent, &sd->groups->flags, &sd_flags_fops); 743 debugfs_create_u32("level", 0444, parent, (u32 *)&sd->level); 744 745 if (sd->flags & SD_ASYM_PACKING) 746 debugfs_create_u32("group_asym_prefer_cpu", 0444, parent, 747 (u32 *)&sd->groups->asym_prefer_cpu); 748 } 749 750 void update_sched_domain_debugfs(void) 751 { 752 int cpu, i; 753 754 /* 755 * This can unfortunately be invoked before sched_debug_init() creates 756 * the debug directory. Don't touch sd_sysctl_cpus until then. 757 */ 758 if (!debugfs_sched) 759 return; 760 761 if (!sched_debug_verbose) 762 return; 763 764 if (!cpumask_available(sd_sysctl_cpus)) { 765 if (!alloc_cpumask_var(&sd_sysctl_cpus, GFP_KERNEL)) 766 return; 767 cpumask_copy(sd_sysctl_cpus, cpu_possible_mask); 768 } 769 770 if (!sd_dentry) { 771 sd_dentry = debugfs_create_dir("domains", debugfs_sched); 772 773 /* rebuild sd_sysctl_cpus if empty since it gets cleared below */ 774 if (cpumask_empty(sd_sysctl_cpus)) 775 cpumask_copy(sd_sysctl_cpus, cpu_online_mask); 776 } 777 778 for_each_cpu(cpu, sd_sysctl_cpus) { 779 struct sched_domain *sd; 780 struct dentry *d_cpu; 781 char buf[32]; 782 783 snprintf(buf, sizeof(buf), "cpu%d", cpu); 784 debugfs_lookup_and_remove(buf, sd_dentry); 785 d_cpu = debugfs_create_dir(buf, sd_dentry); 786 787 i = 0; 788 for_each_domain(cpu, sd) { 789 struct dentry *d_sd; 790 791 snprintf(buf, sizeof(buf), "domain%d", i); 792 d_sd = debugfs_create_dir(buf, d_cpu); 793 794 register_sd(sd, d_sd); 795 i++; 796 } 797 798 __cpumask_clear_cpu(cpu, sd_sysctl_cpus); 799 } 800 } 801 802 void dirty_sched_domain_sysctl(int cpu) 803 { 804 if (cpumask_available(sd_sysctl_cpus)) 805 __cpumask_set_cpu(cpu, sd_sysctl_cpus); 806 } 807 808 #ifdef CONFIG_FAIR_GROUP_SCHED 809 static void print_cfs_group_stats(struct seq_file *m, int cpu, struct task_group *tg) 810 { 811 struct sched_entity *se = tg_se(tg, cpu); 812 813 #define P(F) SEQ_printf(m, " .%-30s: %lld\n", #F, (long long)F) 814 #define P_SCHEDSTAT(F) SEQ_printf(m, " .%-30s: %lld\n", \ 815 #F, (long long)schedstat_val(stats->F)) 816 #define PN(F) SEQ_printf(m, " .%-30s: %lld.%06ld\n", #F, SPLIT_NS((long long)F)) 817 #define PN_SCHEDSTAT(F) SEQ_printf(m, " .%-30s: %lld.%06ld\n", \ 818 #F, SPLIT_NS((long long)schedstat_val(stats->F))) 819 820 if (!se) 821 return; 822 823 PN(se->exec_start); 824 PN(se->vruntime); 825 PN(se->sum_exec_runtime); 826 827 if (schedstat_enabled()) { 828 struct sched_statistics *stats; 829 stats = __schedstats_from_se(se); 830 831 PN_SCHEDSTAT(wait_start); 832 PN_SCHEDSTAT(sleep_start); 833 PN_SCHEDSTAT(block_start); 834 PN_SCHEDSTAT(sleep_max); 835 PN_SCHEDSTAT(block_max); 836 PN_SCHEDSTAT(exec_max); 837 PN_SCHEDSTAT(slice_max); 838 PN_SCHEDSTAT(wait_max); 839 PN_SCHEDSTAT(wait_sum); 840 P_SCHEDSTAT(wait_count); 841 } 842 843 P(se->load.weight); 844 P(se->avg.load_avg); 845 P(se->avg.util_avg); 846 P(se->avg.runnable_avg); 847 848 #undef PN_SCHEDSTAT 849 #undef PN 850 #undef P_SCHEDSTAT 851 #undef P 852 } 853 #endif /* CONFIG_FAIR_GROUP_SCHED */ 854 855 #ifdef CONFIG_CGROUP_SCHED 856 static DEFINE_SPINLOCK(sched_debug_lock); 857 static char group_path[PATH_MAX]; 858 859 static void task_group_path(struct task_group *tg, char *path, int plen) 860 { 861 if (autogroup_path(tg, path, plen)) 862 return; 863 864 cgroup_path(tg->css.cgroup, path, plen); 865 } 866 867 /* 868 * Only 1 SEQ_printf_task_group_path() caller can use the full length 869 * group_path[] for cgroup path. Other simultaneous callers will have 870 * to use a shorter stack buffer. A "..." suffix is appended at the end 871 * of the stack buffer so that it will show up in case the output length 872 * matches the given buffer size to indicate possible path name truncation. 873 */ 874 #define SEQ_printf_task_group_path(m, tg, fmt...) \ 875 { \ 876 if (spin_trylock(&sched_debug_lock)) { \ 877 task_group_path(tg, group_path, sizeof(group_path)); \ 878 SEQ_printf(m, fmt, group_path); \ 879 spin_unlock(&sched_debug_lock); \ 880 } else { \ 881 char buf[128]; \ 882 char *bufend = buf + sizeof(buf) - 3; \ 883 task_group_path(tg, buf, bufend - buf); \ 884 strcpy(bufend - 1, "..."); \ 885 SEQ_printf(m, fmt, buf); \ 886 } \ 887 } 888 #endif 889 890 static void 891 print_task(struct seq_file *m, struct rq *rq, struct task_struct *p) 892 { 893 if (task_current(rq, p)) 894 SEQ_printf(m, ">R"); 895 else 896 SEQ_printf(m, " %c", task_state_to_char(p)); 897 898 SEQ_printf(m, " %15s %5d %9Ld.%06ld %c %9Ld.%06ld %c %9Ld.%06ld %9Ld.%06ld %9Ld %5d ", 899 p->comm, task_pid_nr(p), 900 SPLIT_NS(p->se.vruntime), 901 entity_eligible(cfs_rq_of(&p->se), &p->se) ? 'E' : 'N', 902 SPLIT_NS(p->se.deadline), 903 p->se.custom_slice ? 'S' : ' ', 904 SPLIT_NS(p->se.slice), 905 SPLIT_NS(p->se.sum_exec_runtime), 906 (long long)(p->nvcsw + p->nivcsw), 907 p->prio); 908 909 SEQ_printf(m, "%9lld.%06ld %9lld.%06ld %9lld.%06ld", 910 SPLIT_NS(schedstat_val_or_zero(p->stats.wait_sum)), 911 SPLIT_NS(schedstat_val_or_zero(p->stats.sum_sleep_runtime)), 912 SPLIT_NS(schedstat_val_or_zero(p->stats.sum_block_runtime))); 913 914 #ifdef CONFIG_NUMA_BALANCING 915 SEQ_printf(m, " %d %d", task_node(p), task_numa_group_id(p)); 916 #endif 917 #ifdef CONFIG_CGROUP_SCHED 918 SEQ_printf_task_group_path(m, task_group(p), " %s") 919 #endif 920 921 SEQ_printf(m, "\n"); 922 } 923 924 static void print_rq(struct seq_file *m, struct rq *rq, int rq_cpu) 925 { 926 struct task_struct *g, *p; 927 928 SEQ_printf(m, "\n"); 929 SEQ_printf(m, "runnable tasks:\n"); 930 SEQ_printf(m, " S task PID vruntime eligible " 931 "deadline slice sum-exec switches " 932 "prio wait-time sum-sleep sum-block" 933 #ifdef CONFIG_NUMA_BALANCING 934 " node group-id" 935 #endif 936 #ifdef CONFIG_CGROUP_SCHED 937 " group-path" 938 #endif 939 "\n"); 940 SEQ_printf(m, "-------------------------------------------------------" 941 "------------------------------------------------------" 942 "------------------------------------------------------" 943 #ifdef CONFIG_NUMA_BALANCING 944 "--------------" 945 #endif 946 #ifdef CONFIG_CGROUP_SCHED 947 "--------------" 948 #endif 949 "\n"); 950 951 rcu_read_lock(); 952 for_each_process_thread(g, p) { 953 if (task_cpu(p) != rq_cpu) 954 continue; 955 956 print_task(m, rq, p); 957 } 958 rcu_read_unlock(); 959 } 960 961 void print_cfs_rq(struct seq_file *m, int cpu, struct cfs_rq *cfs_rq) 962 { 963 s64 left_vruntime = -1, right_vruntime = -1, left_deadline = -1, spread; 964 s64 zero_vruntime = -1, sum_w_vruntime = -1; 965 u64 avruntime; 966 struct sched_entity *last, *first, *root; 967 struct rq *rq = cpu_rq(cpu); 968 unsigned int sum_shift; 969 unsigned long flags; 970 u64 sum_weight; 971 972 #ifdef CONFIG_FAIR_GROUP_SCHED 973 SEQ_printf(m, "\n"); 974 SEQ_printf_task_group_path(m, cfs_rq->tg, "cfs_rq[%d]:%s\n", cpu); 975 #else 976 SEQ_printf(m, "\n"); 977 SEQ_printf(m, "cfs_rq[%d]:\n", cpu); 978 #endif 979 980 raw_spin_rq_lock_irqsave(rq, flags); 981 root = __pick_root_entity(cfs_rq); 982 if (root) 983 left_vruntime = root->min_vruntime; 984 first = __pick_first_entity(cfs_rq); 985 if (first) 986 left_deadline = first->deadline; 987 last = __pick_last_entity(cfs_rq); 988 if (last) 989 right_vruntime = last->vruntime; 990 zero_vruntime = cfs_rq->zero_vruntime; 991 sum_w_vruntime = cfs_rq->sum_w_vruntime; 992 sum_weight = cfs_rq->sum_weight; 993 sum_shift = cfs_rq->sum_shift; 994 avruntime = avg_vruntime(cfs_rq); 995 raw_spin_rq_unlock_irqrestore(rq, flags); 996 997 SEQ_printf(m, " .%-30s: %Ld.%06ld\n", "left_deadline", 998 SPLIT_NS(left_deadline)); 999 SEQ_printf(m, " .%-30s: %Ld.%06ld\n", "left_vruntime", 1000 SPLIT_NS(left_vruntime)); 1001 SEQ_printf(m, " .%-30s: %Ld.%06ld\n", "zero_vruntime", 1002 SPLIT_NS(zero_vruntime)); 1003 SEQ_printf(m, " .%-30s: %Ld (%d bits)\n", "sum_w_vruntime", 1004 sum_w_vruntime, ilog2(abs(sum_w_vruntime))); 1005 SEQ_printf(m, " .%-30s: %Lu\n", "sum_weight", 1006 sum_weight); 1007 SEQ_printf(m, " .%-30s: %u\n", "sum_shift", sum_shift); 1008 SEQ_printf(m, " .%-30s: %Ld.%06ld\n", "avg_vruntime", 1009 SPLIT_NS(avruntime)); 1010 SEQ_printf(m, " .%-30s: %Ld.%06ld\n", "right_vruntime", 1011 SPLIT_NS(right_vruntime)); 1012 spread = right_vruntime - left_vruntime; 1013 SEQ_printf(m, " .%-30s: %Ld.%06ld\n", "spread", SPLIT_NS(spread)); 1014 SEQ_printf(m, " .%-30s: %d\n", "nr_queued", cfs_rq->nr_queued); 1015 SEQ_printf(m, " .%-30s: %d\n", "h_nr_runnable", cfs_rq->h_nr_runnable); 1016 SEQ_printf(m, " .%-30s: %d\n", "h_nr_queued", cfs_rq->h_nr_queued); 1017 SEQ_printf(m, " .%-30s: %d\n", "h_nr_idle", cfs_rq->h_nr_idle); 1018 SEQ_printf(m, " .%-30s: %ld\n", "load", cfs_rq->load.weight); 1019 SEQ_printf(m, " .%-30s: %lu\n", "load_avg", 1020 cfs_rq->avg.load_avg); 1021 SEQ_printf(m, " .%-30s: %lu\n", "runnable_avg", 1022 cfs_rq->avg.runnable_avg); 1023 SEQ_printf(m, " .%-30s: %lu\n", "util_avg", 1024 cfs_rq->avg.util_avg); 1025 SEQ_printf(m, " .%-30s: %u\n", "util_est", 1026 cfs_rq->avg.util_est); 1027 SEQ_printf(m, " .%-30s: %ld\n", "removed.load_avg", 1028 cfs_rq->removed.load_avg); 1029 SEQ_printf(m, " .%-30s: %ld\n", "removed.util_avg", 1030 cfs_rq->removed.util_avg); 1031 SEQ_printf(m, " .%-30s: %ld\n", "removed.runnable_avg", 1032 cfs_rq->removed.runnable_avg); 1033 #ifdef CONFIG_FAIR_GROUP_SCHED 1034 SEQ_printf(m, " .%-30s: %lu\n", "tg_load_avg_contrib", 1035 cfs_rq->tg_load_avg_contrib); 1036 SEQ_printf(m, " .%-30s: %ld\n", "tg_load_avg", 1037 atomic_long_read(&cfs_rq->tg->load_avg)); 1038 #endif /* CONFIG_FAIR_GROUP_SCHED */ 1039 #ifdef CONFIG_CFS_BANDWIDTH 1040 SEQ_printf(m, " .%-30s: %d\n", "throttled", 1041 cfs_rq->throttled); 1042 SEQ_printf(m, " .%-30s: %d\n", "throttle_count", 1043 cfs_rq->throttle_count); 1044 #endif 1045 1046 #ifdef CONFIG_FAIR_GROUP_SCHED 1047 print_cfs_group_stats(m, cpu, cfs_rq->tg); 1048 #endif 1049 } 1050 1051 void print_rt_rq(struct seq_file *m, int cpu, struct rt_rq *rt_rq) 1052 { 1053 #ifdef CONFIG_RT_GROUP_SCHED 1054 SEQ_printf(m, "\n"); 1055 SEQ_printf_task_group_path(m, rt_rq->tg, "rt_rq[%d]:%s\n", cpu); 1056 #else 1057 SEQ_printf(m, "\n"); 1058 SEQ_printf(m, "rt_rq[%d]:\n", cpu); 1059 #endif 1060 1061 #define P(x) \ 1062 SEQ_printf(m, " .%-30s: %Ld\n", #x, (long long)(rt_rq->x)) 1063 #define PU(x) \ 1064 SEQ_printf(m, " .%-30s: %lu\n", #x, (unsigned long)(rt_rq->x)) 1065 #define PN(x) \ 1066 SEQ_printf(m, " .%-30s: %Ld.%06ld\n", #x, SPLIT_NS(rt_rq->x)) 1067 1068 PU(rt_nr_running); 1069 1070 #ifdef CONFIG_RT_GROUP_SCHED 1071 P(rt_throttled); 1072 PN(rt_time); 1073 PN(rt_runtime); 1074 #endif 1075 1076 #undef PN 1077 #undef PU 1078 #undef P 1079 } 1080 1081 void print_dl_rq(struct seq_file *m, int cpu, struct dl_rq *dl_rq) 1082 { 1083 struct dl_bw *dl_bw; 1084 1085 SEQ_printf(m, "\n"); 1086 SEQ_printf(m, "dl_rq[%d]:\n", cpu); 1087 1088 #define PU(x) \ 1089 SEQ_printf(m, " .%-30s: %lu\n", #x, (unsigned long)(dl_rq->x)) 1090 1091 PU(dl_nr_running); 1092 dl_bw = &cpu_rq(cpu)->rd->dl_bw; 1093 SEQ_printf(m, " .%-30s: %lld\n", "dl_bw->bw", dl_bw->bw); 1094 SEQ_printf(m, " .%-30s: %lld\n", "dl_bw->total_bw", dl_bw->total_bw); 1095 1096 #undef PU 1097 } 1098 1099 static void print_cpu(struct seq_file *m, int cpu) 1100 { 1101 struct rq *rq = cpu_rq(cpu); 1102 1103 #ifdef CONFIG_X86 1104 { 1105 unsigned int freq = cpu_khz ? : 1; 1106 1107 SEQ_printf(m, "cpu#%d, %u.%03u MHz\n", 1108 cpu, freq / 1000, (freq % 1000)); 1109 } 1110 #else /* !CONFIG_X86: */ 1111 SEQ_printf(m, "cpu#%d\n", cpu); 1112 #endif /* !CONFIG_X86 */ 1113 1114 #define P(x) \ 1115 do { \ 1116 if (sizeof(rq->x) == 4) \ 1117 SEQ_printf(m, " .%-30s: %d\n", #x, (int)(rq->x)); \ 1118 else \ 1119 SEQ_printf(m, " .%-30s: %Ld\n", #x, (long long)(rq->x));\ 1120 } while (0) 1121 1122 #define PN(x) \ 1123 SEQ_printf(m, " .%-30s: %Ld.%06ld\n", #x, SPLIT_NS(rq->x)) 1124 1125 P(nr_running); 1126 P(nr_switches); 1127 P(nr_uninterruptible); 1128 PN(next_balance); 1129 SEQ_printf(m, " .%-30s: %ld\n", "curr->pid", (long)(task_pid_nr(rq->curr))); 1130 PN(clock); 1131 PN(clock_task); 1132 #undef P 1133 #undef PN 1134 1135 #define P64(n) SEQ_printf(m, " .%-30s: %Ld\n", #n, rq->n); 1136 P64(avg_idle); 1137 P64(max_idle_balance_cost); 1138 #undef P64 1139 1140 #define P(n) SEQ_printf(m, " .%-30s: %d\n", #n, schedstat_val(rq->n)); 1141 if (schedstat_enabled()) { 1142 P(yld_count); 1143 P(sched_count); 1144 P(sched_goidle); 1145 P(ttwu_count); 1146 P(ttwu_local); 1147 } 1148 #undef P 1149 1150 print_cfs_stats(m, cpu); 1151 print_rt_stats(m, cpu); 1152 print_dl_stats(m, cpu); 1153 1154 print_rq(m, rq, cpu); 1155 SEQ_printf(m, "\n"); 1156 } 1157 1158 static const char *sched_tunable_scaling_names[] = { 1159 "none", 1160 "logarithmic", 1161 "linear" 1162 }; 1163 1164 static void sched_debug_header(struct seq_file *m) 1165 { 1166 u64 ktime, sched_clk, cpu_clk; 1167 unsigned long flags; 1168 1169 local_irq_save(flags); 1170 ktime = ktime_to_ns(ktime_get()); 1171 sched_clk = sched_clock(); 1172 cpu_clk = local_clock(); 1173 local_irq_restore(flags); 1174 1175 SEQ_printf(m, "Sched Debug Version: v0.11, %s %.*s\n", 1176 init_utsname()->release, 1177 (int)strcspn(init_utsname()->version, " "), 1178 init_utsname()->version); 1179 1180 #define P(x) \ 1181 SEQ_printf(m, "%-40s: %Ld\n", #x, (long long)(x)) 1182 #define PN(x) \ 1183 SEQ_printf(m, "%-40s: %Ld.%06ld\n", #x, SPLIT_NS(x)) 1184 PN(ktime); 1185 PN(sched_clk); 1186 PN(cpu_clk); 1187 P(jiffies); 1188 #ifdef CONFIG_HAVE_UNSTABLE_SCHED_CLOCK 1189 P(sched_clock_stable()); 1190 #endif 1191 #undef PN 1192 #undef P 1193 1194 SEQ_printf(m, "\n"); 1195 SEQ_printf(m, "sysctl_sched\n"); 1196 1197 #define P(x) \ 1198 SEQ_printf(m, " .%-40s: %Ld\n", #x, (long long)(x)) 1199 #define PN(x) \ 1200 SEQ_printf(m, " .%-40s: %Ld.%06ld\n", #x, SPLIT_NS(x)) 1201 PN(sysctl_sched_base_slice); 1202 P(sysctl_sched_features); 1203 #undef PN 1204 #undef P 1205 1206 SEQ_printf(m, " .%-40s: %d (%s)\n", 1207 "sysctl_sched_tunable_scaling", 1208 sysctl_sched_tunable_scaling, 1209 sched_tunable_scaling_names[sysctl_sched_tunable_scaling]); 1210 SEQ_printf(m, "\n"); 1211 } 1212 1213 static int sched_debug_show(struct seq_file *m, void *v) 1214 { 1215 int cpu = (unsigned long)(v - 2); 1216 1217 if (cpu != -1) 1218 print_cpu(m, cpu); 1219 else 1220 sched_debug_header(m); 1221 1222 return 0; 1223 } 1224 1225 void sysrq_sched_debug_show(void) 1226 { 1227 int cpu; 1228 1229 sched_debug_header(NULL); 1230 for_each_online_cpu(cpu) { 1231 /* 1232 * Need to reset softlockup watchdogs on all CPUs, because 1233 * another CPU might be blocked waiting for us to process 1234 * an IPI or stop_machine. 1235 */ 1236 touch_nmi_watchdog(); 1237 touch_all_softlockup_watchdogs(); 1238 print_cpu(NULL, cpu); 1239 } 1240 } 1241 1242 /* 1243 * This iterator needs some explanation. 1244 * It returns 1 for the header position. 1245 * This means 2 is CPU 0. 1246 * In a hotplugged system some CPUs, including CPU 0, may be missing so we have 1247 * to use cpumask_* to iterate over the CPUs. 1248 */ 1249 static void *sched_debug_start(struct seq_file *file, loff_t *offset) 1250 { 1251 unsigned long n = *offset; 1252 1253 if (n == 0) 1254 return (void *) 1; 1255 1256 n--; 1257 1258 if (n > 0) 1259 n = cpumask_next(n - 1, cpu_online_mask); 1260 else 1261 n = cpumask_first(cpu_online_mask); 1262 1263 *offset = n + 1; 1264 1265 if (n < nr_cpu_ids) 1266 return (void *)(unsigned long)(n + 2); 1267 1268 return NULL; 1269 } 1270 1271 static void *sched_debug_next(struct seq_file *file, void *data, loff_t *offset) 1272 { 1273 (*offset)++; 1274 return sched_debug_start(file, offset); 1275 } 1276 1277 static void sched_debug_stop(struct seq_file *file, void *data) 1278 { 1279 } 1280 1281 static const struct seq_operations sched_debug_sops = { 1282 .start = sched_debug_start, 1283 .next = sched_debug_next, 1284 .stop = sched_debug_stop, 1285 .show = sched_debug_show, 1286 }; 1287 1288 #define __PS(S, F) SEQ_printf(m, "%-45s:%21Ld\n", S, (long long)(F)) 1289 #define __P(F) __PS(#F, F) 1290 #define P(F) __PS(#F, p->F) 1291 #define PM(F, M) __PS(#F, p->F & (M)) 1292 #define __PSN(S, F) SEQ_printf(m, "%-45s:%14Ld.%06ld\n", S, SPLIT_NS((long long)(F))) 1293 #define __PN(F) __PSN(#F, F) 1294 #define PN(F) __PSN(#F, p->F) 1295 1296 1297 #ifdef CONFIG_NUMA_BALANCING 1298 void print_numa_stats(struct seq_file *m, int node, unsigned long tsf, 1299 unsigned long tpf, unsigned long gsf, unsigned long gpf) 1300 { 1301 SEQ_printf(m, "numa_faults node=%d ", node); 1302 SEQ_printf(m, "task_private=%lu task_shared=%lu ", tpf, tsf); 1303 SEQ_printf(m, "group_private=%lu group_shared=%lu\n", gpf, gsf); 1304 } 1305 #endif 1306 1307 1308 static void sched_show_numa(struct task_struct *p, struct seq_file *m) 1309 { 1310 #ifdef CONFIG_NUMA_BALANCING 1311 if (p->mm) 1312 P(mm->numa_scan_seq); 1313 1314 P(numa_pages_migrated); 1315 P(numa_preferred_nid); 1316 P(total_numa_faults); 1317 SEQ_printf(m, "current_node=%d, numa_group_id=%d\n", 1318 task_node(p), task_numa_group_id(p)); 1319 show_numa_stats(p, m); 1320 #endif /* CONFIG_NUMA_BALANCING */ 1321 } 1322 1323 void proc_sched_show_task(struct task_struct *p, struct pid_namespace *ns, 1324 struct seq_file *m) 1325 { 1326 unsigned long nr_switches; 1327 1328 SEQ_printf(m, "%s (%d, #threads: %d)\n", p->comm, task_pid_nr_ns(p, ns), 1329 get_nr_threads(p)); 1330 SEQ_printf(m, 1331 "---------------------------------------------------------" 1332 "----------\n"); 1333 1334 #define P_SCHEDSTAT(F) __PS(#F, schedstat_val(p->stats.F)) 1335 #define PN_SCHEDSTAT(F) __PSN(#F, schedstat_val(p->stats.F)) 1336 1337 PN(se.exec_start); 1338 PN(se.vruntime); 1339 PN(se.sum_exec_runtime); 1340 1341 nr_switches = p->nvcsw + p->nivcsw; 1342 1343 P(se.nr_migrations); 1344 1345 if (schedstat_enabled()) { 1346 u64 avg_atom, avg_per_cpu; 1347 1348 PN_SCHEDSTAT(sum_sleep_runtime); 1349 PN_SCHEDSTAT(sum_block_runtime); 1350 PN_SCHEDSTAT(wait_start); 1351 PN_SCHEDSTAT(sleep_start); 1352 PN_SCHEDSTAT(block_start); 1353 PN_SCHEDSTAT(sleep_max); 1354 PN_SCHEDSTAT(block_max); 1355 PN_SCHEDSTAT(exec_max); 1356 PN_SCHEDSTAT(slice_max); 1357 PN_SCHEDSTAT(wait_max); 1358 PN_SCHEDSTAT(wait_sum); 1359 P_SCHEDSTAT(wait_count); 1360 PN_SCHEDSTAT(iowait_sum); 1361 P_SCHEDSTAT(iowait_count); 1362 P_SCHEDSTAT(nr_migrations_cold); 1363 P_SCHEDSTAT(nr_failed_migrations_affine); 1364 P_SCHEDSTAT(nr_failed_migrations_running); 1365 P_SCHEDSTAT(nr_failed_migrations_hot); 1366 P_SCHEDSTAT(nr_forced_migrations); 1367 P_SCHEDSTAT(nr_wakeups); 1368 P_SCHEDSTAT(nr_wakeups_sync); 1369 P_SCHEDSTAT(nr_wakeups_migrate); 1370 P_SCHEDSTAT(nr_wakeups_local); 1371 P_SCHEDSTAT(nr_wakeups_remote); 1372 P_SCHEDSTAT(nr_wakeups_affine); 1373 P_SCHEDSTAT(nr_wakeups_affine_attempts); 1374 P_SCHEDSTAT(nr_wakeups_passive); 1375 P_SCHEDSTAT(nr_wakeups_idle); 1376 1377 avg_atom = p->se.sum_exec_runtime; 1378 if (nr_switches) 1379 avg_atom = div64_ul(avg_atom, nr_switches); 1380 else 1381 avg_atom = -1LL; 1382 1383 avg_per_cpu = p->se.sum_exec_runtime; 1384 if (p->se.nr_migrations) { 1385 avg_per_cpu = div64_u64(avg_per_cpu, 1386 p->se.nr_migrations); 1387 } else { 1388 avg_per_cpu = -1LL; 1389 } 1390 1391 __PN(avg_atom); 1392 __PN(avg_per_cpu); 1393 1394 #ifdef CONFIG_SCHED_CORE 1395 PN_SCHEDSTAT(core_forceidle_sum); 1396 #endif 1397 } 1398 1399 __P(nr_switches); 1400 __PS("nr_voluntary_switches", p->nvcsw); 1401 __PS("nr_involuntary_switches", p->nivcsw); 1402 1403 P(se.load.weight); 1404 P(se.avg.load_sum); 1405 P(se.avg.runnable_sum); 1406 P(se.avg.util_sum); 1407 P(se.avg.load_avg); 1408 P(se.avg.runnable_avg); 1409 P(se.avg.util_avg); 1410 P(se.avg.last_update_time); 1411 PM(se.avg.util_est, ~UTIL_AVG_UNCHANGED); 1412 #ifdef CONFIG_UCLAMP_TASK 1413 __PS("uclamp.min", p->uclamp_req[UCLAMP_MIN].value); 1414 __PS("uclamp.max", p->uclamp_req[UCLAMP_MAX].value); 1415 __PS("effective uclamp.min", uclamp_eff_value(p, UCLAMP_MIN)); 1416 __PS("effective uclamp.max", uclamp_eff_value(p, UCLAMP_MAX)); 1417 #endif /* CONFIG_UCLAMP_TASK */ 1418 P(policy); 1419 P(prio); 1420 if (task_has_dl_policy(p)) { 1421 P(dl.runtime); 1422 P(dl.deadline); 1423 } else if (fair_policy(p->policy)) { 1424 P(se.slice); 1425 } 1426 #ifdef CONFIG_SCHED_CLASS_EXT 1427 __PS("ext.enabled", task_on_scx(p)); 1428 #endif 1429 #undef PN_SCHEDSTAT 1430 #undef P_SCHEDSTAT 1431 1432 { 1433 unsigned int this_cpu = raw_smp_processor_id(); 1434 u64 t0, t1; 1435 1436 t0 = cpu_clock(this_cpu); 1437 t1 = cpu_clock(this_cpu); 1438 __PS("clock-delta", t1-t0); 1439 } 1440 1441 sched_show_numa(p, m); 1442 } 1443 1444 void proc_sched_set_task(struct task_struct *p) 1445 { 1446 #ifdef CONFIG_SCHEDSTATS 1447 memset(&p->stats, 0, sizeof(p->stats)); 1448 #endif 1449 } 1450 1451 void resched_latency_warn(int cpu, u64 latency) 1452 { 1453 static DEFINE_RATELIMIT_STATE(latency_check_ratelimit, 60 * 60 * HZ, 1); 1454 1455 if (likely(!__ratelimit(&latency_check_ratelimit))) 1456 return; 1457 1458 pr_err("sched: CPU %d need_resched set for > %llu ns (%d ticks) without schedule\n", 1459 cpu, latency, cpu_rq(cpu)->ticks_without_resched); 1460 dump_stack(); 1461 } 1462